Identifying Somatic Variants With Defined Confidence Level In Targeted Sequencing Of Tumor Dna Without A Matched Normal Sample

Sequencing tumor DNA without a matched normal sample is currently routine practice for many clinical diagnostic services that use NGS to identify somatic mutations in a panel of cancer-relevant genes, due to cost constraints. Various computational algorithms have been developed to distinguish somatic mutations from germline variations in this “single sample” setting. However, the sensitivity and specificity of these methods have not been rigorously investigated, and factors affecting accuracy of the somatic classification have not been thoroughly mapped out. We seek to address these inadequacies and identify avenues towards better somatic variant identification in “single sample” setting. We use tumor/matched normal pairs to establish true somatic/germline status of variants and analyze how adjustments in the algorithms for variant calling and somatic/germline classification affect the ability to identify true somatic variants. We show that somatic variants can be identified with varying degrees of certainty under different circumstances and demonstrate that our algorithmic methods can effectively identify a substantial subset of calls that have a very high likelihood of being truly somatic. Tunable filtering methods and confidence scoring for somatic likelihood has utility in the development of complex biomarkers such as tumor mutational burden and ctDNA measurements. Citation Format: Yanmei Huang, Amanda G. Young, Jason D. Hughes. Identifying somatic variants with defined confidence level in targeted sequencing of tumor DNA without a matched normal sample [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 5455.